xref: /netbsd-src/sys/dev/raidframe/rf_states.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: rf_states.c,v 1.35 2004/03/23 13:09:18 oster Exp $	*/
2 /*
3  * Copyright (c) 1995 Carnegie-Mellon University.
4  * All rights reserved.
5  *
6  * Author: Mark Holland, William V. Courtright II, Robby Findler
7  *
8  * Permission to use, copy, modify and distribute this software and
9  * its documentation is hereby granted, provided that both the copyright
10  * notice and this permission notice appear in all copies of the
11  * software, derivative works or modified versions, and any portions
12  * thereof, and that both notices appear in supporting documentation.
13  *
14  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
15  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
16  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
17  *
18  * Carnegie Mellon requests users of this software to return to
19  *
20  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
21  *  School of Computer Science
22  *  Carnegie Mellon University
23  *  Pittsburgh PA 15213-3890
24  *
25  * any improvements or extensions that they make and grant Carnegie the
26  * rights to redistribute these changes.
27  */
28 
29 #include <sys/cdefs.h>
30 __KERNEL_RCSID(0, "$NetBSD: rf_states.c,v 1.35 2004/03/23 13:09:18 oster Exp $");
31 
32 #include <sys/errno.h>
33 
34 #include "rf_archs.h"
35 #include "rf_threadstuff.h"
36 #include "rf_raid.h"
37 #include "rf_dag.h"
38 #include "rf_desc.h"
39 #include "rf_aselect.h"
40 #include "rf_general.h"
41 #include "rf_states.h"
42 #include "rf_dagutils.h"
43 #include "rf_driver.h"
44 #include "rf_engine.h"
45 #include "rf_map.h"
46 #include "rf_etimer.h"
47 #include "rf_kintf.h"
48 
49 #ifndef RF_DEBUG_STATES
50 #define RF_DEBUG_STATES 0
51 #endif
52 
53 /* prototypes for some of the available states.
54 
55    States must:
56 
57      - not block.
58 
59      - either schedule rf_ContinueRaidAccess as a callback and return
60        RF_TRUE, or complete all of their work and return RF_FALSE.
61 
62      - increment desc->state when they have finished their work.
63 */
64 
65 #if RF_DEBUG_STATES
66 static char *
67 StateName(RF_AccessState_t state)
68 {
69 	switch (state) {
70 		case rf_QuiesceState:return "QuiesceState";
71 	case rf_MapState:
72 		return "MapState";
73 	case rf_LockState:
74 		return "LockState";
75 	case rf_CreateDAGState:
76 		return "CreateDAGState";
77 	case rf_ExecuteDAGState:
78 		return "ExecuteDAGState";
79 	case rf_ProcessDAGState:
80 		return "ProcessDAGState";
81 	case rf_CleanupState:
82 		return "CleanupState";
83 	case rf_LastState:
84 		return "LastState";
85 	case rf_IncrAccessesCountState:
86 		return "IncrAccessesCountState";
87 	case rf_DecrAccessesCountState:
88 		return "DecrAccessesCountState";
89 	default:
90 		return "!!! UnnamedState !!!";
91 	}
92 }
93 #endif
94 
95 void
96 rf_ContinueRaidAccess(RF_RaidAccessDesc_t *desc)
97 {
98 	int     suspended = RF_FALSE;
99 	int     current_state_index = desc->state;
100 	RF_AccessState_t current_state = desc->states[current_state_index];
101 #if RF_DEBUG_STATES
102 	int     unit = desc->raidPtr->raidid;
103 #endif
104 
105 	do {
106 
107 		current_state_index = desc->state;
108 		current_state = desc->states[current_state_index];
109 
110 		switch (current_state) {
111 
112 		case rf_QuiesceState:
113 			suspended = rf_State_Quiesce(desc);
114 			break;
115 		case rf_IncrAccessesCountState:
116 			suspended = rf_State_IncrAccessCount(desc);
117 			break;
118 		case rf_MapState:
119 			suspended = rf_State_Map(desc);
120 			break;
121 		case rf_LockState:
122 			suspended = rf_State_Lock(desc);
123 			break;
124 		case rf_CreateDAGState:
125 			suspended = rf_State_CreateDAG(desc);
126 			break;
127 		case rf_ExecuteDAGState:
128 			suspended = rf_State_ExecuteDAG(desc);
129 			break;
130 		case rf_ProcessDAGState:
131 			suspended = rf_State_ProcessDAG(desc);
132 			break;
133 		case rf_CleanupState:
134 			suspended = rf_State_Cleanup(desc);
135 			break;
136 		case rf_DecrAccessesCountState:
137 			suspended = rf_State_DecrAccessCount(desc);
138 			break;
139 		case rf_LastState:
140 			suspended = rf_State_LastState(desc);
141 			break;
142 		}
143 
144 		/* after this point, we cannot dereference desc since
145 		 * desc may have been freed. desc is only freed in
146 		 * LastState, so if we renter this function or loop
147 		 * back up, desc should be valid. */
148 
149 #if RF_DEBUG_STATES
150 		if (rf_printStatesDebug) {
151 			printf("raid%d: State: %-24s StateIndex: %3i desc: 0x%ld %s\n",
152 			       unit, StateName(current_state),
153 			       current_state_index, (long) desc,
154 			       suspended ? "callback scheduled" : "looping");
155 		}
156 #endif
157 	} while (!suspended && current_state != rf_LastState);
158 
159 	return;
160 }
161 
162 
163 void
164 rf_ContinueDagAccess(RF_DagList_t *dagList)
165 {
166 #if RF_ACC_TRACE > 0
167 	RF_AccTraceEntry_t *tracerec = &(dagList->desc->tracerec);
168 	RF_Etimer_t timer;
169 #endif
170 	RF_RaidAccessDesc_t *desc;
171 	RF_DagHeader_t *dag_h;
172 	int     i;
173 
174 	desc = dagList->desc;
175 
176 #if RF_ACC_TRACE > 0
177 	timer = tracerec->timer;
178 	RF_ETIMER_STOP(timer);
179 	RF_ETIMER_EVAL(timer);
180 	tracerec->specific.user.exec_us = RF_ETIMER_VAL_US(timer);
181 	RF_ETIMER_START(tracerec->timer);
182 #endif
183 
184 	/* skip to dag which just finished */
185 	dag_h = dagList->dags;
186 	for (i = 0; i < dagList->numDagsDone; i++) {
187 		dag_h = dag_h->next;
188 	}
189 
190 	/* check to see if retry is required */
191 	if (dag_h->status == rf_rollBackward) {
192 		/* when a dag fails, mark desc status as bad and allow
193 		 * all other dags in the desc to execute to
194 		 * completion.  then, free all dags and start over */
195 		desc->status = 1;	/* bad status */
196 #if 0
197 		printf("raid%d: DAG failure: %c addr 0x%lx "
198 		       "(%ld) nblk 0x%x (%d) buf 0x%lx state %d\n",
199 		       desc->raidPtr->raidid, desc->type,
200 		       (long) desc->raidAddress,
201 		       (long) desc->raidAddress, (int) desc->numBlocks,
202 		       (int) desc->numBlocks,
203 		       (unsigned long) (desc->bufPtr), desc->state);
204 #endif
205 	}
206 	dagList->numDagsDone++;
207 	rf_ContinueRaidAccess(desc);
208 }
209 
210 int
211 rf_State_LastState(RF_RaidAccessDesc_t *desc)
212 {
213 	void    (*callbackFunc) (RF_CBParam_t) = desc->callbackFunc;
214 	RF_CBParam_t callbackArg;
215 
216 	callbackArg.p = desc->callbackArg;
217 
218 	/*
219 	 * If this is not an async request, wake up the caller
220 	 */
221 	if (desc->async_flag == 0)
222 		wakeup(desc->bp);
223 
224 	/*
225 	 * That's all the IO for this one... unbusy the 'disk'.
226 	 */
227 
228 	rf_disk_unbusy(desc);
229 
230 	/*
231 	 * Wakeup any requests waiting to go.
232 	 */
233 
234 	RF_LOCK_MUTEX(((RF_Raid_t *) desc->raidPtr)->mutex);
235 	((RF_Raid_t *) desc->raidPtr)->openings++;
236 	RF_UNLOCK_MUTEX(((RF_Raid_t *) desc->raidPtr)->mutex);
237 
238 	/* wake up any pending IO */
239 	raidstart(((RF_Raid_t *) desc->raidPtr));
240 
241 	/* printf("Calling biodone on 0x%x\n",desc->bp); */
242 	biodone(desc->bp);	/* access came through ioctl */
243 
244 	if (callbackFunc)
245 		callbackFunc(callbackArg);
246 	rf_FreeRaidAccDesc(desc);
247 
248 	return RF_FALSE;
249 }
250 
251 int
252 rf_State_IncrAccessCount(RF_RaidAccessDesc_t *desc)
253 {
254 	RF_Raid_t *raidPtr;
255 
256 	raidPtr = desc->raidPtr;
257 	/* Bummer. We have to do this to be 100% safe w.r.t. the increment
258 	 * below */
259 	RF_LOCK_MUTEX(raidPtr->access_suspend_mutex);
260 	raidPtr->accs_in_flight++;	/* used to detect quiescence */
261 	RF_UNLOCK_MUTEX(raidPtr->access_suspend_mutex);
262 
263 	desc->state++;
264 	return RF_FALSE;
265 }
266 
267 int
268 rf_State_DecrAccessCount(RF_RaidAccessDesc_t *desc)
269 {
270 	RF_Raid_t *raidPtr;
271 
272 	raidPtr = desc->raidPtr;
273 
274 	RF_LOCK_MUTEX(raidPtr->access_suspend_mutex);
275 	raidPtr->accs_in_flight--;
276 	if (raidPtr->accesses_suspended && raidPtr->accs_in_flight == 0) {
277 		rf_SignalQuiescenceLock(raidPtr);
278 	}
279 	RF_UNLOCK_MUTEX(raidPtr->access_suspend_mutex);
280 
281 	desc->state++;
282 	return RF_FALSE;
283 }
284 
285 int
286 rf_State_Quiesce(RF_RaidAccessDesc_t *desc)
287 {
288 #if RF_ACC_TRACE > 0
289 	RF_AccTraceEntry_t *tracerec = &desc->tracerec;
290 	RF_Etimer_t timer;
291 #endif
292 	RF_CallbackDesc_t *cb;
293 	RF_Raid_t *raidPtr;
294 	int     suspended = RF_FALSE;
295 	int need_cb, used_cb;
296 
297 	raidPtr = desc->raidPtr;
298 
299 #if RF_ACC_TRACE > 0
300 	RF_ETIMER_START(timer);
301 	RF_ETIMER_START(desc->timer);
302 #endif
303 
304 	need_cb = 0;
305 	used_cb = 0;
306 	cb = NULL;
307 
308 	RF_LOCK_MUTEX(raidPtr->access_suspend_mutex);
309 	/* Do an initial check to see if we might need a callback structure */
310 	if (raidPtr->accesses_suspended) {
311 		need_cb = 1;
312 	}
313 	RF_UNLOCK_MUTEX(raidPtr->access_suspend_mutex);
314 
315 	if (need_cb) {
316 		/* create a callback if we might need it...
317 		   and we likely do. */
318 		cb = rf_AllocCallbackDesc();
319 	}
320 
321 	RF_LOCK_MUTEX(raidPtr->access_suspend_mutex);
322 	if (raidPtr->accesses_suspended) {
323 		cb->callbackFunc = (void (*) (RF_CBParam_t)) rf_ContinueRaidAccess;
324 		cb->callbackArg.p = (void *) desc;
325 		cb->next = raidPtr->quiesce_wait_list;
326 		raidPtr->quiesce_wait_list = cb;
327 		suspended = RF_TRUE;
328 		used_cb = 1;
329 	}
330 	RF_UNLOCK_MUTEX(raidPtr->access_suspend_mutex);
331 
332 	if ((need_cb == 1) && (used_cb == 0)) {
333 		rf_FreeCallbackDesc(cb);
334 	}
335 
336 #if RF_ACC_TRACE > 0
337 	RF_ETIMER_STOP(timer);
338 	RF_ETIMER_EVAL(timer);
339 	tracerec->specific.user.suspend_ovhd_us += RF_ETIMER_VAL_US(timer);
340 #endif
341 
342 #if RF_DEBUG_QUIESCE
343 	if (suspended && rf_quiesceDebug)
344 		printf("Stalling access due to quiescence lock\n");
345 #endif
346 	desc->state++;
347 	return suspended;
348 }
349 
350 int
351 rf_State_Map(RF_RaidAccessDesc_t *desc)
352 {
353 	RF_Raid_t *raidPtr = desc->raidPtr;
354 #if RF_ACC_TRACE > 0
355 	RF_AccTraceEntry_t *tracerec = &desc->tracerec;
356 	RF_Etimer_t timer;
357 
358 	RF_ETIMER_START(timer);
359 #endif
360 
361 	if (!(desc->asmap = rf_MapAccess(raidPtr, desc->raidAddress, desc->numBlocks,
362 		    desc->bufPtr, RF_DONT_REMAP)))
363 		RF_PANIC();
364 
365 #if RF_ACC_TRACE > 0
366 	RF_ETIMER_STOP(timer);
367 	RF_ETIMER_EVAL(timer);
368 	tracerec->specific.user.map_us = RF_ETIMER_VAL_US(timer);
369 #endif
370 
371 	desc->state++;
372 	return RF_FALSE;
373 }
374 
375 int
376 rf_State_Lock(RF_RaidAccessDesc_t *desc)
377 {
378 #if RF_ACC_TRACE > 0
379 	RF_AccTraceEntry_t *tracerec = &desc->tracerec;
380 	RF_Etimer_t timer;
381 #endif
382 	RF_Raid_t *raidPtr = desc->raidPtr;
383 	RF_AccessStripeMapHeader_t *asmh = desc->asmap;
384 	RF_AccessStripeMap_t *asm_p;
385 	RF_StripeNum_t lastStripeID = -1;
386 	int     suspended = RF_FALSE;
387 
388 #if RF_ACC_TRACE > 0
389 	RF_ETIMER_START(timer);
390 #endif
391 
392 	/* acquire each lock that we don't already hold */
393 	for (asm_p = asmh->stripeMap; asm_p; asm_p = asm_p->next) {
394 		RF_ASSERT(RF_IO_IS_R_OR_W(desc->type));
395 		if (!rf_suppressLocksAndLargeWrites &&
396 		    asm_p->parityInfo &&
397 		    !(desc->flags & RF_DAG_SUPPRESS_LOCKS) &&
398 		    !(asm_p->flags & RF_ASM_FLAGS_LOCK_TRIED)) {
399 			asm_p->flags |= RF_ASM_FLAGS_LOCK_TRIED;
400 				/* locks must be acquired hierarchically */
401 			RF_ASSERT(asm_p->stripeID > lastStripeID);
402 			lastStripeID = asm_p->stripeID;
403 
404 			RF_INIT_LOCK_REQ_DESC(asm_p->lockReqDesc, desc->type,
405 					      (void (*) (struct buf *)) rf_ContinueRaidAccess, desc, asm_p,
406 					      raidPtr->Layout.dataSectorsPerStripe);
407 			if (rf_AcquireStripeLock(raidPtr->lockTable, asm_p->stripeID,
408 						 &asm_p->lockReqDesc)) {
409 				suspended = RF_TRUE;
410 				break;
411 			}
412 		}
413 		if (desc->type == RF_IO_TYPE_WRITE &&
414 		    raidPtr->status == rf_rs_reconstructing) {
415 			if (!(asm_p->flags & RF_ASM_FLAGS_FORCE_TRIED)) {
416 				int     val;
417 
418 				asm_p->flags |= RF_ASM_FLAGS_FORCE_TRIED;
419 				val = rf_ForceOrBlockRecon(raidPtr, asm_p,
420 							   (void (*) (RF_Raid_t *, void *)) rf_ContinueRaidAccess, desc);
421 				if (val == 0) {
422 					asm_p->flags |= RF_ASM_FLAGS_RECON_BLOCKED;
423 				} else {
424 					suspended = RF_TRUE;
425 					break;
426 				}
427 			} else {
428 #if RF_DEBUG_PSS > 0
429 				if (rf_pssDebug) {
430 					printf("raid%d: skipping force/block because already done, psid %ld\n",
431 					       desc->raidPtr->raidid,
432 					       (long) asm_p->stripeID);
433 				}
434 #endif
435 			}
436 		} else {
437 #if RF_DEBUG_PSS > 0
438 			if (rf_pssDebug) {
439 				printf("raid%d: skipping force/block because not write or not under recon, psid %ld\n",
440 				       desc->raidPtr->raidid,
441 				       (long) asm_p->stripeID);
442 			}
443 #endif
444 		}
445 	}
446 #if RF_ACC_TRACE > 0
447 	RF_ETIMER_STOP(timer);
448 	RF_ETIMER_EVAL(timer);
449 	tracerec->specific.user.lock_us += RF_ETIMER_VAL_US(timer);
450 #endif
451 	if (suspended)
452 		return (RF_TRUE);
453 
454 	desc->state++;
455 	return (RF_FALSE);
456 }
457 /*
458  * the following three states create, execute, and post-process dags
459  * the error recovery unit is a single dag.
460  * by default, SelectAlgorithm creates an array of dags, one per parity stripe
461  * in some tricky cases, multiple dags per stripe are created
462  *   - dags within a parity stripe are executed sequentially (arbitrary order)
463  *   - dags for distinct parity stripes are executed concurrently
464  *
465  * repeat until all dags complete successfully -or- dag selection fails
466  *
467  * while !done
468  *   create dag(s) (SelectAlgorithm)
469  *   if dag
470  *     execute dag (DispatchDAG)
471  *     if dag successful
472  *       done (SUCCESS)
473  *     else
474  *       !done (RETRY - start over with new dags)
475  *   else
476  *     done (FAIL)
477  */
478 int
479 rf_State_CreateDAG(RF_RaidAccessDesc_t *desc)
480 {
481 #if RF_ACC_TRACE > 0
482 	RF_AccTraceEntry_t *tracerec = &desc->tracerec;
483 	RF_Etimer_t timer;
484 #endif
485 	RF_DagHeader_t *dag_h;
486 	RF_DagList_t *dagList;
487 	struct buf *bp;
488 	int     i, selectStatus;
489 
490 	/* generate a dag for the access, and fire it off.  When the dag
491 	 * completes, we'll get re-invoked in the next state. */
492 #if RF_ACC_TRACE > 0
493 	RF_ETIMER_START(timer);
494 #endif
495 	/* SelectAlgorithm returns one or more dags */
496 	selectStatus = rf_SelectAlgorithm(desc, desc->flags | RF_DAG_SUPPRESS_LOCKS);
497 #if RF_DEBUG_VALIDATE_DAG
498 	if (rf_printDAGsDebug) {
499 		dagList = desc->dagList;
500 		for (i = 0; i < desc->numStripes; i++) {
501 			rf_PrintDAGList(dagList.dags);
502 			dagList = dagList->next;
503 		}
504 	}
505 #endif /* RF_DEBUG_VALIDATE_DAG */
506 #if RF_ACC_TRACE > 0
507 	RF_ETIMER_STOP(timer);
508 	RF_ETIMER_EVAL(timer);
509 	/* update time to create all dags */
510 	tracerec->specific.user.dag_create_us = RF_ETIMER_VAL_US(timer);
511 #endif
512 
513 	desc->status = 0;	/* good status */
514 
515 	if (selectStatus) {
516 		/* failed to create a dag */
517 		/* this happens when there are too many faults or incomplete
518 		 * dag libraries */
519 		printf("raid%d: failed to create a dag. "
520 		       "Too many component failures.\n",
521 		       desc->raidPtr->raidid);
522 
523 		desc->status = 1; /* bad status */
524 		/* skip straight to rf_State_Cleanup() */
525 		desc->state = rf_CleanupState;
526 		bp = (struct buf *)desc->bp;
527 		bp->b_flags |= B_ERROR;
528 		bp->b_error = EIO;
529 	} else {
530 		/* bind dags to desc */
531 		dagList = desc->dagList;
532 		for (i = 0; i < desc->numStripes; i++) {
533 			dag_h = dagList->dags;
534 			while (dag_h) {
535 				dag_h->bp = (struct buf *) desc->bp;
536 #if RF_ACC_TRACE > 0
537 				dag_h->tracerec = tracerec;
538 #endif
539 				dag_h = dag_h->next;
540 			}
541 			dagList = dagList->next;
542 		}
543 		desc->flags |= RF_DAG_DISPATCH_RETURNED;
544 		desc->state++;	/* next state should be rf_State_ExecuteDAG */
545 	}
546 	return RF_FALSE;
547 }
548 
549 
550 
551 /* the access has an list of dagLists, one dagList per parity stripe.
552  * fire the first dag in each parity stripe (dagList).
553  * dags within a stripe (dagList) must be executed sequentially
554  *  - this preserves atomic parity update
555  * dags for independents parity groups (stripes) are fired concurrently */
556 
557 int
558 rf_State_ExecuteDAG(RF_RaidAccessDesc_t *desc)
559 {
560 	int     i;
561 	RF_DagHeader_t *dag_h;
562 	RF_DagList_t *dagList;
563 
564 	/* next state is always rf_State_ProcessDAG important to do
565 	 * this before firing the first dag (it may finish before we
566 	 * leave this routine) */
567 	desc->state++;
568 
569 	/* sweep dag array, a stripe at a time, firing the first dag
570 	 * in each stripe */
571 	dagList = desc->dagList;
572 	for (i = 0; i < desc->numStripes; i++) {
573 		RF_ASSERT(dagList->numDags > 0);
574 		RF_ASSERT(dagList->numDagsDone == 0);
575 		RF_ASSERT(dagList->numDagsFired == 0);
576 #if RF_ACC_TRACE > 0
577 		RF_ETIMER_START(dagList->tracerec.timer);
578 #endif
579 		/* fire first dag in this stripe */
580 		dag_h = dagList->dags;
581 		RF_ASSERT(dag_h);
582 		dagList->numDagsFired++;
583 		rf_DispatchDAG(dag_h, (void (*) (void *)) rf_ContinueDagAccess, dagList);
584 		dagList = dagList->next;
585 	}
586 
587 	/* the DAG will always call the callback, even if there was no
588 	 * blocking, so we are always suspended in this state */
589 	return RF_TRUE;
590 }
591 
592 
593 
594 /* rf_State_ProcessDAG is entered when a dag completes.
595  * first, check to all dags in the access have completed
596  * if not, fire as many dags as possible */
597 
598 int
599 rf_State_ProcessDAG(RF_RaidAccessDesc_t *desc)
600 {
601 	RF_AccessStripeMapHeader_t *asmh = desc->asmap;
602 	RF_Raid_t *raidPtr = desc->raidPtr;
603 	RF_DagHeader_t *dag_h;
604 	int     i, j, done = RF_TRUE;
605 	RF_DagList_t *dagList, *temp;
606 
607 	/* check to see if this is the last dag */
608 	dagList = desc->dagList;
609 	for (i = 0; i < desc->numStripes; i++) {
610 		if (dagList->numDags != dagList->numDagsDone)
611 			done = RF_FALSE;
612 		dagList = dagList->next;
613 	}
614 
615 	if (done) {
616 		if (desc->status) {
617 			/* a dag failed, retry */
618 			/* free all dags */
619 			dagList = desc->dagList;
620 			for (i = 0; i < desc->numStripes; i++) {
621 				rf_FreeDAG(dagList->dags);
622 				temp = dagList;
623 				dagList = dagList->next;
624 				rf_FreeDAGList(temp);
625 			}
626 			rf_MarkFailuresInASMList(raidPtr, asmh);
627 			/* back up to rf_State_CreateDAG */
628 			desc->state = desc->state - 2;
629 			return RF_FALSE;
630 		} else {
631 			/* move on to rf_State_Cleanup */
632 			desc->state++;
633 		}
634 		return RF_FALSE;
635 	} else {
636 		/* more dags to execute */
637 		/* see if any are ready to be fired.  if so, fire them */
638 		/* don't fire the initial dag in a list, it's fired in
639 		 * rf_State_ExecuteDAG */
640 		dagList = desc->dagList;
641 		for (i = 0; i < desc->numStripes; i++) {
642 			if ((dagList->numDagsDone < dagList->numDags)
643 			    && (dagList->numDagsDone == dagList->numDagsFired)
644 			    && (dagList->numDagsFired > 0)) {
645 #if RF_ACC_TRACE > 0
646 				RF_ETIMER_START(dagList->tracerec.timer);
647 #endif
648 				/* fire next dag in this stripe */
649 				/* first, skip to next dag awaiting execution */
650 				dag_h = dagList->dags;
651 				for (j = 0; j < dagList->numDagsDone; j++)
652 					dag_h = dag_h->next;
653 				dagList->numDagsFired++;
654 				rf_DispatchDAG(dag_h, (void (*) (void *)) rf_ContinueDagAccess,
655 				    dagList);
656 			}
657 			dagList = dagList->next;
658 		}
659 		return RF_TRUE;
660 	}
661 }
662 /* only make it this far if all dags complete successfully */
663 int
664 rf_State_Cleanup(RF_RaidAccessDesc_t *desc)
665 {
666 #if RF_ACC_TRACE > 0
667 	RF_AccTraceEntry_t *tracerec = &desc->tracerec;
668 	RF_Etimer_t timer;
669 #endif
670 	RF_AccessStripeMapHeader_t *asmh = desc->asmap;
671 	RF_Raid_t *raidPtr = desc->raidPtr;
672 	RF_AccessStripeMap_t *asm_p;
673 	RF_DagList_t *dagList;
674 	int i;
675 
676 	desc->state++;
677 
678 #if RF_ACC_TRACE > 0
679 	timer = tracerec->timer;
680 	RF_ETIMER_STOP(timer);
681 	RF_ETIMER_EVAL(timer);
682 	tracerec->specific.user.dag_retry_us = RF_ETIMER_VAL_US(timer);
683 
684 	/* the RAID I/O is complete.  Clean up. */
685 	tracerec->specific.user.dag_retry_us = 0;
686 
687 	RF_ETIMER_START(timer);
688 #endif
689 	/* free all dags */
690 	dagList = desc->dagList;
691 	for (i = 0; i < desc->numStripes; i++) {
692 		rf_FreeDAG(dagList->dags);
693 		dagList = dagList->next;
694 	}
695 #if RF_ACC_TRACE > 0
696 	RF_ETIMER_STOP(timer);
697 	RF_ETIMER_EVAL(timer);
698 	tracerec->specific.user.cleanup_us = RF_ETIMER_VAL_US(timer);
699 
700 	RF_ETIMER_START(timer);
701 #endif
702 	for (asm_p = asmh->stripeMap; asm_p; asm_p = asm_p->next) {
703 		if (!rf_suppressLocksAndLargeWrites &&
704 		    asm_p->parityInfo &&
705 		    !(desc->flags & RF_DAG_SUPPRESS_LOCKS)) {
706 			RF_ASSERT_VALID_LOCKREQ(&asm_p->lockReqDesc);
707 			rf_ReleaseStripeLock(raidPtr->lockTable,
708 					     asm_p->stripeID,
709 					     &asm_p->lockReqDesc);
710 		}
711 		if (asm_p->flags & RF_ASM_FLAGS_RECON_BLOCKED) {
712 			rf_UnblockRecon(raidPtr, asm_p);
713 		}
714 	}
715 #if RF_ACC_TRACE > 0
716 	RF_ETIMER_STOP(timer);
717 	RF_ETIMER_EVAL(timer);
718 	tracerec->specific.user.lock_us += RF_ETIMER_VAL_US(timer);
719 
720 	RF_ETIMER_START(timer);
721 #endif
722 	rf_FreeAccessStripeMap(asmh);
723 #if RF_ACC_TRACE > 0
724 	RF_ETIMER_STOP(timer);
725 	RF_ETIMER_EVAL(timer);
726 	tracerec->specific.user.cleanup_us += RF_ETIMER_VAL_US(timer);
727 
728 	RF_ETIMER_STOP(desc->timer);
729 	RF_ETIMER_EVAL(desc->timer);
730 
731 	timer = desc->tracerec.tot_timer;
732 	RF_ETIMER_STOP(timer);
733 	RF_ETIMER_EVAL(timer);
734 	desc->tracerec.total_us = RF_ETIMER_VAL_US(timer);
735 
736 	rf_LogTraceRec(raidPtr, tracerec);
737 #endif
738 	desc->flags |= RF_DAG_ACCESS_COMPLETE;
739 
740 	return RF_FALSE;
741 }
742